Specific embodiment
The embodiment of present disclosure is directed to the method and system of the time series Transformation Analysis for data.For example,Data sample can be the sensing data from semiconductor processing device.In one embodiment, this method and system can be examinedSurvey the probability that new time series data matches previous time series data.It can be by using the k arest neighbors in embodiment(k-Nearest Neighbor, kNN) analysis and logistic regression (logistic regression, LR) Lai Zhihang time seriesTransformation Analysis.The embodiment of present disclosure is extensible in terms of the susceptibility of adjustable time series Transformation Analysis's.
As technique (such as manufacturing process) includes the steps that shorter and shorter time, smaller and smaller component, increasingly tighterTolerance of lattice etc., transformation (such as how to be become more and more important in a manufacturing process from step A to step B).If technique turnsBecome overshoot (overshoot) or less stress (undershoot) and (such as too fast 20 degree is converted to from 10 degree, too slowly from 10 degree turnsChange to 20 degree etc.), then problem may occur.Repeatable performance includes consistent transformation.Traditional monitoring method (such asSPC instantaneous time sequence) can not be monitored and can not be detected and migrated over time from the received data of sensor (referred to herein asSensor time sequence data) in short time signal disturbance.These short exceptions may cause defect (such as on chip lackFall into) or reduction yield.
Time series Transformation Analysis provides the ability of monitoring period sequence transitions.Time series Transformation Analysis is detectable notCan be detected by conventional method rare, strange and/or not expected sequence (such as time series data (is directed toThe value that sample is drawn) the shape of curve, numerical value, position etc.；Referring to Fig. 4).In one embodiment, by from going throughHistory time series data estimates expected transitional locus and by the track of new time series data and historical time sequence dataTrack be compared, Lai Zhihang time series transformation monitoring.Time series Transformation Analysis also can detect short exception and mentionThe accuracy carried out sensitization for tuner parameters or sensitization is gone to detect.Time series Transformation Analysis can also overcome the mistake of conventional methodReport rate.For example, guard band analysis may provide wrong report, the wrong report is that entire signal is protected because the least part of signal is located atProtect frequency band except without being matched with echo signal, and time series Transformation Analysis provide Signal Matching echo signal probability andWrong report is not provided.In another embodiment, time series Transformation Analysis can be used to detect short time signal disturbance and (such as captureDisturbance mark (such as search for similarity)) to search all examples of FDC.
The fault detection classification (fault detection classification, FDC) of time series data can monitorData from single-sensor, the data may make classification inaccurate.Can by monitor at any time one change it is moreA signal come extract more information (such as before pressure spike valve location variation may indicate that asking on pressure control logicThe problem of topic, the pressure spike before valve location variation may indicate that pressure sensor etc.).Technology disclosed hereinHandle a coupled signal to change at any time.
Time series Transformation Analysis can be by k arest neighbors (kNN) method (such as kNN algorithm) and logistic regression (LR) binary pointClass device combines the monitoring to realize time series.The combination of kNN and LR can be used specific inclined on detection time sequence dataFrom (excursion).Time series Transformation Analysis can be used kNN by every time window (such as 1 on 100 seconds time intervalsSecond time slip-window) short period sequence transitions be simplified to single dimension to determine the distance away from anticipatory behavior.Time sequenceLR can be used to establish binary classifier in column Transformation Analysis, and the binary classifier, which is used to generate new time series data, to be hadOr without target pattern (pattern) probability (such as new time series data whether by kNN method determined away fromFrom except).
Time series Transformation Analysis can be used to based on turn between the set point change in time series data characterization processesBecome, and detects the deviation relative to expected transitional locus in new time series data.Expected transitional locus can by whenBetween sequence data defined.
Fig. 1 is painted the network architecture 100 according to an embodiment.Originally, time series Transformation Analysis system 102 identifiesData source 106A-N (such as sensor), data source 106A-N define system and/or are used to monitoring system (such as entity handles systemSystem is 104).Entity handles system 104 can be semiconductor processing device, such as chamber, deposition chambers for etch reactor etc.Deng.User can via client machine 110 from the various data sources in data source 106A-N (such as via graphical user interface(GUI)) time series data (such as sample) is selected.Time series Transformation Analysis system 102 generates training dataset and is based onTraining dataset and time series data calculate distance value.
In one embodiment, user can also select deviation 108 (that is, anomalous system behavior via client machine 110Defined parameter), and deviate and 108 can be stored in permanent storage unit 112 by time series Transformation Analysis system 102.
For example, entity handles system 104 may include manufacture tool or by directly or via network (such as local area network(LAN)) it is connected to manufacture tool.The example of manufacture tool includes the semiconductor manufacturing tool for manufacturing electronic equipment, such asEtcher, chemical vapor deposition stove etc..The step of manufacturing such equipment may include being related to the number of different types of manufacturing processTen manufacturing steps, these steps can be described as being formulated.
Entity handles system 104 may include any kind of calculating equipment (including desktop computer, laptop computer,Programmable logic controller (PLC) (PLC), handheld computer or similar calculating equipment) carry out control system.Data source 106 (such asSensor) can for entity handles system 104 and/or manufacture tool a part, or may be connected to entity handles system 104 and/Or manufacture tool (such as via network).
Client machine 110 can be any kind of calculating equipment, including desktop computer, laptop computer, movementCommunication equipment, mobile phone, smart phone, handheld computer or similar calculating equipment.
In one embodiment, entity handles system 104, data source 106, permanent storage unit 112 and client machinesDevice 110 is connected to time series Transformation Analysis system 102, can be and is directly connected to or via hardware interface (not shown) or warpIt is indirectly connected with by what network (not shown) carried out.Network can be local area network (LAN) (such as internal network in company), wirelessNetwork, mobile communications network or wide area network (WAN) (such as internet or similar communication system).Network may include any quantityNetworking and calculate equipment, such as wired and wireless device.
Function division presented above is only through exemplary mode to carry out.In other embodiments, describedFunction is combined into monolithic element or is subdivided into any component combination.For example, client machine 110 and time sequence transitionsAnalysis system 102 can be hosted in single computer system, in individual computer system or combinations thereof on.
Fig. 2 is painted an embodiment of the method 200 for time series Transformation Analysis.It can be held by processing logicRow method 200, the processing logic may include hardware (such as circuit, special logic, programmable logic, microcode etc.), softPart (such as run instruction) on a processing device or combinations thereof.In one embodiment, method 200 is the time by Fig. 1Sequence transitions analysis system 102 is performed.
At the square 202 of Fig. 2, the processing logic receiving time sequence data 402 of time series Transformation Analysis system 102(such as echo signal), as shown in Figure 4.One or more sensors can generate the time during technique (such as manufacturing process)Sequence data 402.Time series data 402 may include a data point more than first.A data point may include time series more than firstData point at the sample of data 402.For example, as shown in Figure 4, can at n=25 and n+1=50 intercepted samples.Time sequenceThe value of column data 402 may include the t (n) and t (n+1) about at [0,4].
Fig. 2 is returned to, at square 204, it includes randomization number that the processing logic of time series Transformation Analysis system 102, which generates,The training dataset at strong point 502 (such as random sample), as shown in Figure 5 A.Randomization data point 502 may include for oppositeIn one or more desired extent distributions deviateed of time series data 402.The distribution can be normal distribution or another pointCloth.In one embodiment, 100 random samples are generated, wherein each random sample is indicated relative to time series numberAccording to deviation.As shown in the example in Fig. 5 A, randomization data point 502 includes each (such as n and n+1) in data pointThe deviation at place.For example, data point is had accumulated around [0,4] at n=25 and n=50.When randomization data point 502 can be used asBetween sequence data 402 pattern training set.When each randomization data point in randomization data point 502 can correspond to come fromBetween sequence data 402 one of more than first a data points.
Fig. 2 is returned to, at square 206, the processing logic use of time series Transformation Analysis system 102 is in time window 506The interior set of randomization data point 502 combines to generate randomization data point, as shown in Figure 5 A.For example, randomization data pointCombination may include one of randomization data point 502a in the example of the time window 506 of from 0 to n (such as 25) and from n to nOne of randomization data point 502b in the example of the time window 506 of+1 (such as 25 to 50).In one embodiment,Randomization data point can be generated in the end of time window 506 by handling logic (referring for example to Fig. 5 A).In another embodiment,Randomization data point can be generated in the midpoint of time window 506 by handling logic.In another embodiment, processing logic can whenBetween window 506 beginning generate randomization data point.
Time window 506 can be time slip-window, and technique can occur in the certain time section for being greater than time slip-window.Time slip-window can be for from currently in time toward the period stretched in the past.For example, two seconds sliding windows may include having occurredAny sample or data point in past two seconds.In an embodiment of time slip-window, the first example can be 0-25, theTwo examples can be 25-50 etc..Therefore, window was slided with 25 seconds.In another embodiment of time slip-window, firstExample can also be 0-25, and the second example can be 1-26, then 2-27 etc..Therefore, time window was with 1 second (or other times unit)It is slided.
The generation of randomization data point combination can be executed for each in multiple examples of time slip-window 506.It is describedEach example in multiple examples may span across the different periods in time interval (for example, the combination of randomization data point includes from nFirst data point at place and the sample of the second data point at n+1).
Fig. 2 is returned to, at square 208, the processing logic of time series Transformation Analysis system 102 is based on randomization data pointCombination is to calculate distance value.First distance value can be calculated for the combination of the first randomization data point.First distance value can indicate moreFirst subset of the set of a randomization data point is away from the combined distance of the first subset of more than first a data points.It can be for slidingEach in multiple examples of time window executes the calculating of distance value.
As shown in Figure 5 B, randomization data point can be combined to provide randomization data point combination 507, randomization dataPoint combination 507 respectively includes the first randomization data point from t (n) and the second randomization data point from t (n+1).ThisA little randomization data point combinations 507 can be used to calculate distance value using k nearest neighbor algorithm.
K arest neighbors (kNN) algorithm can be used to be directed to each example calculation distance threshold of time window 506 in processing logic.For example, first distance threshold value can be generated for the time window 506 at time t=25 (for example, using the data at time 0-25Point), second distance threshold value (for example, using the data point at time 1-26) can be generated for the time window 506 at time t=26Etc..The calculating of distance threshold may include calculating the randomization data for each in multiple randomization data points combination 507Point combines the Euclidean distance between 507 and each remaining randomization data point combination 507 from training dataset (referring to figure5B).The calculating of distance threshold may include identifying the smallest Euclidean distance from the Euclidean distance being computed.The smallest Euclidean distance canFor distance threshold.
Using the algorithm of kNN type, training dataset can be used to estimate bias sample and training data itBetween distance.For each training sample for including the combination of randomization data point, can calculate all in this sample and training setEuclidean distance between other samples, and k-th of minimum value can be stored.It is by equation d for sample jj=smallk(xj-X distance) is calculated, wherein X is the matrix of n × m.The quantity (such as 100 random samples) of value n expression training sample.The value of mThe quantity of time samples or data point can be indicated (for example, at two of n=25 and n+1=50 depicted in the example of Fig. 4-6Between sample).Variable xjFor m element vector (such as [0,4]) and it can indicate j-th in X column.For all in training setSample repeats this processing and generates the neighborhood or limit vector L with n element.Neighborhood or limit vector L can be used to generate goodThe training set separated well trains simple classifier.The random sample of self-training collection can be elected to calculate for each sampleknn=smallk(xj-X).The random sample for being not from training set can be selected and in order to which visual purpose calculates knnValue.
As shown in Figure 6, random sample (sample class 602a) is had selected from deviation pattern, and has estimated knnMeasurement.ChoosingGo out the random sample (sample class 602b) for showing non-deviation behavior, and has estimated knnMeasurement.Sample class 602a is shownAway from training set compared with the smaller distance of sample class 602b.In Fig. 6,2D signal by simplifying for seem can linear separation oneDimension amount.
It has been directed to the above-mentioned processing of the pattern representation including two data points.However, this identical processing can be expandedTo various dimensions simplifying multidimensional input for single measurement (such as k-th of distance between sample and training data).It is being directed toWhen all probable values examine Fig. 6, a minimum value is at the about from deviating from the position of [0,4].Fig. 7 A-7B is painted multiple input patternsKNN measurement and the minimum value being presented at about [0,4] is shown.
Fig. 2 is returned to, at square 210, the processing logic of time series Transformation Analysis system 102 is based on the distance being computedIt is worth next life constituent class device.Handling logic can be by based on multiple range estimation distance threshold next life constituent class device being computed.It canThe generation of classifier is executed for each in multiple examples of time slip-window 506.Logistic regression next life constituent class can be usedDevice.
802 (logistic fits (logit fit) can be returned from training data decision logic by handling logic；As shown in Figure 8)(for example, generating the logistic fit 802 for being directed to training data, the probability of Signal Matching deviation will be generated).Training data can wrapInclude primordial time series data and the combination of randomization data point and their distance value being computed.Equation p (y | X)=1/ (1+e-β*X) it can be used to decision logic recurrence 802.Training data is used to estimate β.As shown in Figure 8, logistic regression 802 may include fromOne position of the transformation pattern of the first data point (sample class 602a) to the second data point (sample class 602b).Transformation figureCase can reflect near the reflection point 804 being centrally positioned on transformation pattern.Time series data 402 can be detected as having and closeThe jump function (such as scalariform deposition of short step) of key transformation.Time series Transformation Analysis can be used to overcome via boundaryIt is reported by mistake caused by method.
Time series transient analysis can utilize tuner parameters.The controllable sample pair beyond specification of time series Transformation AnalysisHow many contributed apart from.The contribution for increasing the sample beyond specification will be so that system be more sensitive.Increase reflection point 804 and makes systemIt is less sensitive.The slope of adjustment logistic regression 802 changes the probability of the sample close to reflection 804.Logistic regression 802 can haveIt reflects the limit (for example, vertical line), and can be considered as mismatching expected behavior more than any sample of the reflection limit.At oneIn embodiment, compared to shallower or less shallow transformation pattern may be needed for transformation pattern shown in fig. 8.θ is availableIt will change as tuner parameters and be adjusted to shallower or less shallow.
Fig. 9 depicts the logistic regression 802 with the θ 902 for being adjusted to generate shallower transformation.Turn using shallowerIn the case where change, all input t can be directed tonAnd tn+1Estimated probability.In the case where being estimated using β, probability can be in Fig. 7 A-7BIt is maximized at middle determined minimum value.
Processing logic can receive the first parameter (such as θ 902) to adjust the susceptibility of the judgement of probability.For example, θ 902a canWith one value, θ 902b can have two value, and θ 902c can have five value.Handling logic can be adjusted based on the first parameter 902The either shallow of the whole transformation pattern around reflection point 804.Tuning knob can be used to variation θ 902 and tuning be set as muting sensitive senseDegree, high sensitive etc..
Fig. 2 is returned to, at square 212, the processing logic of time series Transformation Analysis system 102 is determined using classifierThe probability of new time series data matching primordial time series data.Processing logic can receive new time series data and meterThe second distance value between new time series in primordial time series data in evaluation time window 506 and time window 506.PlaceWhether reason logic classifier the can be used new time series data to determine in time window 506 has the more than distance thresholdTwo distance values, and in response to the new time series data in time window 506 be more than distance threshold judgement and generate failure orNotice.
Figure 10 A-10B is directed to all values [tn,tn+1] show input to the probability of match time sequence data 402 (for example,There is maximum value at [0,4]).
Figure 11 A-11D is painted various new time series data 1002 in the case where using time series Transformation AnalysisProbability with time series data 402.In Figure 11 A, new time series data 1102a has substantive match time sequence numberAccording to the pattern of 402 pattern, and cause about 1 probability.New time series data in Figure 11 B, at n=251102b, which is greater than, to be expected, therefore the probability of match time sequence data 402 is about 0.93.In Figure 11 C, new time series numberIt is higher than time series data 402 at n=25 according to 1102c and is lower than time series data 402 at n=50, therefore matches meshThe probability for marking signal is about 0.5.In Figure 11 D, new time series data 1102d is significant higher at n=25 and in n=50Place is significant lower, therefore the probability for matching target is 0.
Fig. 3 is painted an embodiment of the method 300 for time series Transformation Analysis.It can be held by processing logicRow method 300, the processing logic may include hardware (such as circuit, special logic, programmable logic, microcode etc.), softPart (such as run instruction) on a processing device or combinations thereof.In one embodiment, method 300 is the time by Fig. 1Sequence transitions analysis system 102 is performed.
At square 302, processing logic receives the time series data 1102a-d (reference including more than first a data pointsFigure 11 A-11D).Each more than first in a data point can join from different time correlations.It can be given birth to during technique by sensorAt time series data.
At square 304, processing logic is by more than first from time series data 1102a-d in time window 1106First subset of data point and the second subset of more than second a data points from previous time series data 402 are comparedCompared with.Time window 1106 can be in time from current point in time to the time slip-window of former extension specific quantity.Time slip-windowSpecific quantity can be extended from current point in time to later in time.
At square 306, processing logic calculation indicates the first subset of more than first a data points away from more than second a data pointsSecond subset combined distance distance value.
At square 308, whether processing logic decision distance value is more than distance threshold (1A-11D referring to Fig.1).
At square 310, processing logical response is more than the judgement of distance threshold in distance value and exports notice.In a realityIt applies in mode, notice includes the instruction of the probability of new time series data match time sequence data 402 (for example, being directed toThe 0.996 of new time series data 1102a, for the 0.926 of new time series data 1102b, for new time sequenceThe 0.502 of column data 1102c, for the 0 of new time series data 1102d).In one embodiment, notice includesHave which section time window section of new time series data (for example, correspond to) of new time series data it is new whenBetween sequence data match time sequence data 402 probability threshold value (such as 0.5,0.85) instruction below.Implement at oneIn mode, it can show and notify via graphical user interface (such as via figure, chart, text etc.).In an embodimentIn, notice is one or more of the alarm of sound equipment, vision etc..It in one embodiment, is by phone, electronics postalOne or more of part, text etc. notify to send.In one embodiment, notice output so that tool, device,One or more of component, workshop etc. carry out one or more of following behavior: stopping action, pause activity, slow downActivity, shutdown etc..
Time series Transformation Analysis can be used for abnormality detection.In one embodiment, fault detection and classification (FDC) needleSensing data is formulated to known defect and/or abnormal mark automatic searching.User's setup cost may be low, becauseAnticipatory behavior can be inferred to from historical behavior.The database of known defect can be independently of formula set point.Identical database canAct on multiple formulas.Defect database can be developed in enterprises under controlled environment, and by these defect database portionsIt affixes one's name to scene.Known defect can have the corrective action for allowing quickly to solve known defect.Event can be captured for abnormal markBarrier maintenance knowledge (such as emphasize interested track or sensing data to user, user can mark or abnormal classification mark withAnd addition corrective action).General service condition includes examining for the post-processing formulation data of known defect and about failureIt repairs and the knowledge capture of new defect.
Time series Transformation Analysis can be used for time series deviation detection with for cannot by conventional method (such as SPC, markQuasi- failure monitoring method) detection abnormal behaviour search time sequence.User's setup cost may be low, because of expected rowTo be inferred from historical behavior.Algorithm is designed to intrinsic in tolerance other methods (such as the monitoring of simple guard band)Wrong report.Time series deviation can be stored and be used to search historical data or Future Data.Trouble hunting knowledge can be captured.OneAs service condition include that the post-processing formulation data for known defect, the knowledge about trouble hunting and new defect are caughtIt catches, the analysis of instantaneous time sequence and repeatability are analyzed.
Time series Transformation Analysis can be used to identify problem when technique undergoes mistake.For example, chamber may be undergoingIntermittent pressure spike, but find root the reason of and solution may be difficult, because of one of following reasonOr more persons: lack the data derived from the tool institute, or mistake can not be reappeared in enterprises or at the scene.Between when in useIt, can be for the subset of the history loop-around data of deviation behavior search tools in the case that sequence transitions are analyzed.Deviation row can be foundFor (identifying the multiple strokes for mismatching anticipatory behavior for example, deviateing and searching), the spike in tool data can be matched, and canIt is searched again for deviateing.The generation deviateed several times allows efficiently trouble hunting and solves the problems, such as.Problem can be knownNot Wei specific components function (such as specific valve opening and close pump and cause stress reading on fluctuation).
Time series Transformation Analysis can also be used to detection unstability.For example, tool may use the lower function on formulaRate mark.Candidate formula can be continuously recycled on tool.All strokes of manual analysis may be infeasible, it is thus possible to omitThe problem of low probability and/or frequency of interval.It, can be for time using abnormality detection and time series deviation detectionAll strokes of apolegamy side are for all steps analysis power and the power of reflection.Analysis can quickly identify the row of certain percentageSuspicious actions in journey.Observed certain defects may have potential yield effect.It is sent to the anti-of process exploitation teamFeedback can prompt the change of formula and the repetition of technique.Deviation about 5% can be reduced.
Figure 12 A is painted the time series data for time series Transformation Analysis.As shown in Figure 12, time series data402 n sample or data point is intercepted, rather than 2 samples in time window as shown in Figure 4.In an embodiment partyIt is the intercepted samples at [5,10,15...95] in formula, this measure causes 19 samples, and method 200 or 300 is made to become 19 dimensionsThe problem of spending rather than the problem of 2 dimension.Between when in use in the case where sequence transitions analysis (such as method 200, method 300),Echo signal at each sample point generates training set.
As shown in Figure 12B, random sample (classification 602a) is had selected from deviation pattern, and has estimated knnMeasurement.It selectsShow the random sample (classification 602b) of non-deviation behavior, and has estimated knnMeasurement.As shown in figure 12 c, be using θ be 5To generate logistic regression 802.Using logistic regression 802, various input signal match time sequence numbers can be directed toAccording to the probability for estimating various input signals the case where 402, as shown in Figure 13 A-13D.Time series data 402 is for giving birth toAt the pattern of housebroken classifier.New time series data 1302a-d is additional executes and primordial time series data 402The new signal of associated special process.In figure 13a, new time series data 1302a is relative to time series data 402And it deviates, and matching probability is about 0.6.In Figure 13 B, new time series data 1302b at n=0 to n=50 relative toTime series data 402 is higher, and matching probability is about 0.7.
As shown in Figure 14, time series data 402 may include from first sensor the first data 1402 (such as whenBetween sequence data 1402) and the second data 1404 (such as time series data 1404) from second sensor.Handle logicIt can determine that the time relationship between the first data and the second data (for example, capturing temporarily separating for FDC(temporarily-spaced) covariant (covariate) signal).Each time series data can have on each signalDifferent pattern.In Figure 14, the sinking in time series data 1404 can be associated with the increase in time series data 1402(for example, it may be possible to causing the increase in time series data 1402).Time series Transformation Analysis (such as method 200, method 300)It can be used to the relational pattern of detection time sequence data 1402 and 1404.In one embodiment, by time series data1402 and 1404 connect together in the case where generate single training vector the problem of (for example, generate 39 dimension).Using logicIn the case where recurrence and kNN algorithm, it can estimate that the probability of various input signal match time sequence datas 1402 and 1404 is timelyBetween relationship between sequence data 1402 and 1404.Figure 15 A-D is painted input signal 1502 and 1504 match time sequence datas1402 and 1404 probability.
In one embodiment, two training sets (402 1 training sets of each time series data) are generated and are usedTwo-dimentional logistic regression 802.
In an example, time series Transformation Analysis can receive the data measured by three sensors.Data can wrapInclude forward power data, reflection power data and pressure data.Three signals and their covariance from three sensorsIt can indicate the mark of plasma percussion (plasma strike) deviation.Time series Transformation Analysis can determine that be passed by threeThere is abnormal marks in certain time section in the data that sensor measures.It may be main relative to expected deviationIt is in the forward power signal at about 0.4 second of entry time section.Deviation can cause to be higher than normal reflection power simultaneouslyMark.This can indicate that plasma fires problem.Pressure may show correct shape, but deviate about 0.5 second.Pressure spike canThe label when fired for reflection power.The time series Transformation Analysis of data from three sensors, which can recognize, influences oneOr the abnormal mark of other multiple signal datas starts from where, to determine being what causes plasma percussion deviation.
Figure 16 is block diagram, is painted exemplary computer device (or system) 1600.In one embodiment, equipment is calculated(or system) 1600 can be the time series Transformation Analysis system 102 of Fig. 1.Calculating equipment 1600 includes being used for so that calculating equipment1600 execute the instruction set of any one or more of methodology discussed herein.Machine can be in client-serverWith the functional operation of server machine in device network environment.Machine can for personal computer (PC), set-top box (STB), server,Network router, switch or bridge or any machine for being able to carry out instruction set (in order or otherwise), described instructionThe specified movement to be taken by the machine of collection.Further, though being only painted single calculating equipment, term " calculates equipment "Using come include individually or jointly execute one group (or multiple groups) instruction with execute in methodology discussed herein appointThe machine of one or more any series.
Exemplary computer device 1600 include the processing system (processing equipment) 1602 communicated with each other via bus 1608,Main memory 1604 (such as read-only memory (ROM), flash memory, dynamic random access memory (DRAM) (such as synchronous dram(SDRAM)) etc.), static memory 1606 (such as flash memory, static random access memory (SRAM) etc.) and data storageEquipment 1616.
Processing equipment 1602 indicates one or more general service processing equipments, such as microprocessor, central processing unitEtc..More specifically, processing equipment 1602 can be complex instruction set calculation (complex instruction setComputing, CISC) microprocessor, reduced instruction set computing (reduced instruction set computing,RISC), very long instruction word (very long instruction word, VLIW) microprocessor or implement other instruction setProcessor or the processor for implementing instruction set combination.Processing equipment 1602 can also set for the processing of one or more specific usesIt is standby, such as application-specific integrated circuit (ASIC), field programmable gate array (FPGA), digital signal processor (DSP), networkProcessor etc..Processing equipment 1602 is configured as executing operation and step discussed herein.
Network interface device 1622 can be further included by calculating equipment 1600.Calculating equipment 1600 may also comprise video display unit1610 (such as liquid crystal display (LCD) or cathode-ray tubes (CRT)), literary digital input equipment 1612 (such as keyboard), cursorControl equipment 1614 (such as mouse) and signal generating apparatus 1620 (such as loudspeaker).
Data Holding Equipment 1616 may include that one or more instruction set 1626 are stored on the Data Holding EquipmentComputer-readable storage media 1624, these instruction set realize any of methodology or function described hereinOr more persons.In one embodiment, instruction 1626 includes time series Transformation Analysis system 102.Computer-readable storageMedia 1624 can be the non-transitory computer-readable storage media for including instruction, these instructions are executed by computer systemWhen make computer system execute include time series Transformation Analysis (such as method 200, method 300 etc.) one group of operation.Instruction 1626 (completely or at least partially) can also reside in primary storage during executing these instructions by calculating equipment 1600In device 1604 and/or in processing equipment 1602, main memory 1604 and processing equipment 1602 also constitute computer-readable media.Instruction 1626 can further send or receive on network 1628 via network interface device 1622.
Although computer-readable storage media 1624 is illustrated as single medium, word " meter in the exemplary embodimentCalculation machine readable memory media " application is including that the single medium for storing one or more instruction set or multiple media (such as collectChinese style or distributed data base and/or associated cache and server).Word " computer-readable storage media "Using including that can store, encode or realize any media of the instruction set as performed by machine, and described instruction collection makes machineAny one or more of the methodology of device execution present disclosure.Accordingly, term " computer-readable storage media " is answeredFor including but is not limited to solid-state memory, optical media and magnetic media.
Certain parts of detailed description below are with the symbol of the operation on algorithm and data bit in computer storageNumber indicate meaning present.These algorithmic descriptions and expression be from be familiar with the technical staff of technical field of data processing forThe others skilled in the art of the technical field most effectively convey the means of the substantive content of its work.Algorithm is (and general hereFor) it is contemplated to be the self-congruent sequence of steps for leading to certain result.These steps are the physical manipulations for including entity amountThey's step of behavior.Although in general, may not, this tittle using can be stored, transmit, in conjunction with, compare and otherwiseThe form of the electrical or magnetic signal of manipulation.These signals are censured as position, value, element, symbol, character, item, number etc. sometimesSuitably (the reasons why mainly for general service).
However, should keep firmly in mind, all these terms and similar term are associated with entity amount appropriate and onlyIt is the appropriate label applied to this tittle.Unless otherwise specifically recited, otherwise as understood by discussion below, it is to be understood thatBe to be referred to everywhere in this specification using the discussion of the term of such as " judgement ", " identification ", " comparison ", " transmission " etc.Computer system (or similar electronic computing device) by the buffer and memory by computer system entity (such asElectronics) the represented data manipulation of amount and be transformed by computer system memory or buffer or the storage of other this type of information,Transmission or the movement and program of analogously represented other data of entity amount in display equipment.
The embodiment of present disclosure is also about the system for executing operation herein.It can be directed to described hereinPurpose to come this system of special construction or the system may include general service computer, the general service computer is by storing upThe computer program selective actuation that is stored in computer reconfigures.Such computer program can be stored in computerIn (or machine) readable memory media, such as (but not limited to) any kind of disc (including floppy disk, CD, CD-ROM andMagneto-optic disk), read-only memory (ROM), random access memory (RAM), EPROM, EEPROM, magnetically or optically card, flash memory or be suitable forAny type media of stored electrons instruction.
Algorithms and displays presented herein is substantially not about any certain computer or other devices.Can with according toVarious general purpose systems are used together according to the program of introduction herein, or the building device more becomed privileged executes methodStep may be suitable.The structure of various systems for these systems will show from description herein.In addition, not joiningThe embodiment of present disclosure is described according to any specific program language.It will be appreciated that various program languages can be usedTo implement the teaching of disclosure as described herein.
It is to be understood that described above it is intended that illustrative and not restrictive.Read and understand it is described above itAfterwards, it will be appreciated by persons skilled in the art that many other embodiments.Therefore, the protection scope of present disclosure should refer toThe full scope of the equivalent that appended claims and such claim are assigned determines.